Predicting gamma passing rates for intensity-modulated radiotherapy fields based on deep learning method(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2021年第6期
- Page:
- 677-681
- Research Field:
- 医学放射物理
- Publishing date:
Info
- Title:
- Predicting gamma passing rates for intensity-modulated radiotherapy fields based on deep learning method
- Author(s):
- 1. School of Nuclear Science and Technology; University of South China; Hengyang 421001; China 2. Department of Oncology; Xiangya Hospital; Central South University; Changsha 410008; China
- 1. School of Nuclear Science and Technology, University of South China, Hengyang 421001, China 2. Department of Oncology, Xiangya Hospital, Central South University, Changsha 410008, China
- Keywords:
- Keywords: gliomas deep learning gamma passing rate intensity-modulated radiotherapy quality assurance
- PACS:
- R318；R815.6
- DOI:
- DOI:10.3969/j.issn.1005-202X.2021.06.004
- Abstract:
- Abstract: Objective To develop a convolution neural network (CNN) model for predicting the gamma passing rates (GPR) for intensity-modulated radiotherapy (IMRT) fields. Methods The IMRT plans of 48 gliomas patients were extracted from Eclipse treatment planning system, with a total of 260 radiation fields. The corresponding verification plan of each IMRT plan was designed based on the measurements of electronic portal imaging device, and it was delivered on Varian 23 EX Linac. Then, portal dosimetry system was used for Gamma analysis on calculated dose and actual dose measured by electronic portal imaging device, thereby obtaining GPR under 2% (global)/2 mm criterion. The dose distribution map obtained by portal dosimetry system was taken as input, and the data set was divided into training set (208 fields), validation set (26 fields) and test set (26 fields). A CNN model which was developed based on tensorflow framework was used to learn the correlation between the dose distribution map and GPR and mean absolute error was used to evaluate the prediction effect of the model. Results In validation set and test set, the GPR prediction errors of 96% of samples were less than ±3%, with a maximum prediction error of 3.09% and 3.54% and a mean absolute error of 0.99% and 1.17%. The Pearson correlation coefficients between predicted and measured GPR in validation set and test set were 0.96 and 0.90, respectively. Conclusion The CNN model developed based on deep learning can accurately predict the GPR for gliomas IMRT fields, which can inform the physicists which plans may not meet the requirement of quality assurance in advance, thereby effectively promoting the quality assurance of clinical radiotherapy.
Last Update: 2021-06-29